Article
Automation & Control Systems
Guang Wang, Jinghui Yang, Yucheng Qian, Jingsong Han, Jianfang Jiao
Summary: In this article, a kernel principal component analysis (KPCA)-based canonical correlation analysis (CCA) model is proposed for nonlinear process monitoring in quality-related fault detection and diagnosis (QrFDD). The KPCA is used to eliminate nonlinear coupling among the variables by extracting kernel principal components (KPCs) of original variables data. The KPCs and output are then used for CCA modeling, establishing a linear regression model between process and quality variables based on the proportional relationship between process variables sample and kernel sample under the Gaussian kernel. This nonlinear QrFDD method outperforms existing kernel-based CCA methods in terms of algorithmic complexity and interpretability, as demonstrated by simulation results.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Automation & Control Systems
Yuting Lyu, Le Zhou, Ya Cong, Hongbo Zheng, Zhihuan Song
Summary: A multirate mixture probability principal component analysis model is proposed for process modeling and fault detection in multirate multimode processes. This model can handle multirate data and utilize all available measurements for fault detection and mode identification, even if some variables are unobserved.
IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING
(2023)
Article
Chemistry, Multidisciplinary
Xuemei Wang, Ping Wu
Summary: This paper proposes an improved process monitoring method by considering autocorrelation among process data, integrating ensemble learning and kernel canonical variate analysis. The method achieves significantly enhanced fault detection performance.
Article
Computer Science, Information Systems
Liangliang Shang, Kexin Shi, Chen Ma, Aibing Qiu, Liang Hua
Summary: The proposed method, CV-NPCA, combines explicit second-order polynomial mapping with a new statistic (Qc) for fault detection in non-linear dynamic processes. It outperforms traditional methods such as PCA, CVA, KPCA, NDPCA, and KECA in terms of fault detection and identification rates with lower false alarm rates.
Article
Engineering, Chemical
Hairong Fang, Wenhua Tao, Shan Lu, Zhijiang Lou, Yonghui Wang, Yuanfei Xue
Summary: This paper proposes a new two-step dynamic local kernel principal component analysis method, which can handle the nonlinearity and the dynamic features simultaneously.
Article
Energy & Fuels
Michael Schmid, Jan Kleiner, Christian Endisch
Summary: This study develops a novel data-driven approach to detect internal short circuits (ISCs) in batteries by analyzing the cell voltage differences. By combining multiple kernel functions, fast detection and robust behavior are achieved for various types of ISCs.
JOURNAL OF ENERGY STORAGE
(2022)
Article
Automation & Control Systems
Jingxiang Liu, Dan Wang, Junghui Chen
Summary: A global-local based wavelet functional principal component analysis (WFPCA) method is proposed to continuously analyze process variable trajectories and effectively detect faulty variables in batch processes.
CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS
(2021)
Article
Mathematics, Interdisciplinary Applications
Zhaojing Wang, Weidong Yang, Hong Zhang, Ying Zheng
Summary: This paper introduces a data-driven method for monitoring multimode processes, using statistics pattern analysis to extract statistical information from process data. A support vector data description method is proposed to address nonlinear and non-Gaussian problems, with a modified local reachability density ratio introduced as a weight factor to improve monitoring performance. The effectiveness of the method is demonstrated through example applications.
Article
Engineering, Multidisciplinary
Zhijiang Lou, Zedong Li, Youqing Wang, Shan Lu
Summary: This paper introduces an improved neural component analysis (INCA) method, which addresses the issue of NCA's inability to handle non-Gaussian features by proposing a new cost function based on kurtosis. It also improves the extraction of key information from process data by selecting principal components (PCs) in the original data space. Experimental results show that INCA outperforms other methods in fault detection.
Article
Engineering, Chemical
Simin Li, Shuang-hua Yang, Yi Cao
Summary: Most industrial systems today are nonlinear and dynamic, and traditional fault detection techniques are limited in simultaneously extracting both nonlinear and dynamic features. This work proposes a novel nonlinear dynamic process monitoring method called canonical variate kernel analysis (CVKA), which combines the CVA method for linear dynamic feature extraction and kernel principal component analysis for nonlinear feature extraction. Experimental results on a TE process case study demonstrate the excellent performance of CVKA compared to other common approaches in dynamic nonlinear process monitoring for TE-like processes.
Article
Automation & Control Systems
Min Wang, Donghua Zhou, Maoyin Chen
Summary: In this paper, a model called recursive hybrid variable monitoring (RHVM) is proposed to address the issue of process monitoring with hybrid variables and nonstationarity. RHVM utilizes a recursive strategy to suppress nonstationary trends and reveal fault information, and it has the ability to update itself with arriving samples.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Automation & Control Systems
Donglei Zheng, Le Zhou, Zhihuan Song
Summary: In this paper, a novel kernel multi-rate probabilistic principal component analysis (K-MPPCA) model is proposed to extract nonlinear correlations among different sampling rates. The model parameters are calibrated using the kernel trick and the expectation-maximum (EM) algorithm, and fault detection methods based on nonlinear features are developed. The efficiency of the proposed method is demonstrated through testing in a simulated nonlinear case and an actual pre-decarburization unit in the ammonia synthesis process.
IEEE-CAA JOURNAL OF AUTOMATICA SINICA
(2021)
Article
Engineering, Electrical & Electronic
Qiqi Wu, Xuefeng Yan
Summary: This study proposes a stacked attention autoencoder (SAAE) monitoring model based on the upper and lower bounds of the interval and fault-related variables. Using Jensen-Shannon (JS) divergence as an indicator to measure the difference, variables with significant changes before and after the fault are screened. An attention mechanism (AM) is introduced in the training process of stacked AE (SAE), giving larger weight to features strongly correlated with the screened fault-related variables. This method reduces uncertainty and extracts fault-related information using historical fault data.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Chemical
Miao Mou, Xiaoqiang Zhao
Summary: This paper proposes a method for detecting and diagnosing incipient nonlinear faults with missing data in industrial processes. The method uses low rank matrix decomposition to recover missing data and builds a mixed kernel function model in the recovered data to extract both local information and global characteristics. The dissimilarity statistic is introduced for fault detection. Numerical examples and simulation verification demonstrate the method's good detection and diagnosis capabilities.
JOURNAL OF THE TAIWAN INSTITUTE OF CHEMICAL ENGINEERS
(2022)
Article
Engineering, Chemical
Jian Huang, Xu Yang, Yuri A. W. Shardt, Xuefeng Yan
Summary: A sparse modeling and monitoring approach based on sparse, distributed principal component analysis (SDPCA) is proposed to achieve sparsity in dimensional reduction techniques. By dividing the data set into highly correlated blocks (HCBs) and a remainder block (RB), interpretable principal components are obtained for highly correlated variables and sparsity is achieved for weakly correlated ones. The method also includes a fault diagnosis strategy named blockwise contribution plots, which outperforms PCA and SPCA in detecting faulty samples and providing accurate diagnosis results.
JOURNAL OF THE TAIWAN INSTITUTE OF CHEMICAL ENGINEERS
(2021)
Article
Computer Science, Artificial Intelligence
Tsuyoshi Uchida, Koichi Fujiwara, Kenichi Nishioji, Masao Kobayashi, Manabu Kano, Yuya Seko, Kanji Yamaguchi, Yoshito Itoh, Hiroshi Kadotani
Summary: This study proposes a new method for analyzing the causal relationship in medical checkup data to discover factors of disease progression. By identifying the causal effects of checkup items on disease progression, the underlying mechanisms of diseases can be revealed. The proposed analysis framework can be applied to various medical checkup data and contribute to the discovery of unknown disease factors.
ARTIFICIAL INTELLIGENCE IN MEDICINE
(2022)
Article
Engineering, Chemical
Shota Kato, Sanghong Kim, Masahiko Mizuta, Manabu Kano
Summary: The authors proposed a nonlinear model predictive control method based on the gray-box model and successive linearization for improving the production quality and reducing the cost of silicon ingots in the CZ process. A method for updating the prediction model to handle plant-model mismatch was proposed, and the control simulation results showed that the proposed method outperformed the conventional control method.
JOURNAL OF CHEMICAL ENGINEERING OF JAPAN
(2022)
Article
Clinical Neurology
Ayako Iwasaki, Koichi Fujiwara, Chikao Nakayama, Yukiyoshi Sumi, Manabu Kano, Tetsuharu Nagamoto, Hiroshi Kadotani
Summary: This study validates a SAS screening methodology using R-R interval and long short-term memory technology, achieving high screening performance in a large clinical dataset. The method can contribute to the realization of an easy-to-use SAS screening system.
CLINICAL NEUROPHYSIOLOGY
(2022)
Article
Robotics
Rikumo Ode, Koichi Fujiwara, Miho Miyajima, Toshikata Yamakawa, Manabu Kano, Kazutaka Jin, Nobukazu Nakasato, Yasuko Sawai, Toru Hoshida, Masaki Iwasaki, Yoshiko Murata, Satsuki Watanabe, Yutaka Watanabe, Yoko Suzuki, Motoki Inaji, Naoto Kunii, Satoru Oshino, Hui Ming Khoo, Haruhiko Kishima, Taketoshi Maehara
Summary: This study aims to develop a machine learning algorithm that can predict epileptic seizures in real-time by monitoring R-R interval data. The initial results showed that the algorithm performed well in most patients, with the exception of false positives in specific participants. Further investigation into the causes of false positives and optimization of the algorithm using additional clinical data will be conducted.
ARTIFICIAL LIFE AND ROBOTICS
(2023)
Article
Computer Science, Interdisciplinary Applications
Xinmin Zhang, Manabu Kano, Masahiro Tani
Summary: Soft-sensor SSPAE, a novel data-driven model integrating Poisson regression network layers into the deep autoencoders framework, is proposed. SSPAE can progressively learn quality-related deep features while taking the quality information into account, thereby improving the prediction accuracy. Evaluated with numerical example and real-world data, SSPAE outperforms PLS, SVR, PR, SAE-FCL, and SAE-PR in prediction accuracy.
COMPUTERS & CHEMICAL ENGINEERING
(2023)
Review
Computer Science, Information Systems
Yueyang Luo, Xinmin Zhang, Manabu Kano, Long Deng, Chunjie Yang, Zhihuan Song
Summary: The blast furnace is a highly energy-intensive, highly polluting, and extremely complex reactor in the ironmaking process. Soft sensors play an important role in predicting molten iron quality indices and have attracted increasing attention from researchers. However, there has been no systematic review of data-driven soft sensors in the blast furnace ironmaking process.
FRONTIERS OF INFORMATION TECHNOLOGY & ELECTRONIC ENGINEERING
(2023)
Article
Pharmacology & Pharmacy
Shuichi Tanabe, Tatsuya Muraki, Keita Yaginuma, Sanghong Kim, Manabu Kano
Summary: The implementation of design space is a scientific concept to ensure the quality of drug products for regulatory approval. This study proposes a greedy approach to construct a low-dimensional design space based on a high-dimensional statistical model and observed internal representations, satisfying both comprehensive process understanding and visualization capability.
INTERNATIONAL JOURNAL OF PHARMACEUTICS
(2023)
Review
Automation & Control Systems
Wan Sieng Yeo, Agus Saptoro, Perumal Kumar, Manabu Kano
Summary: Soft sensors are mathematical models that estimate hard-to-measure variables using easy-to-measure variables. Data-driven algorithms are preferred for developing soft sensors due to their profitability and technical feasibility. This paper critically reviews and discusses the existing just-in-time (JIT) based algorithms for developing adaptive soft sensors, highlighting their limitations and considering algorithms for nonlinear and missing data. Recommendations and future directions for JIT-based algorithms are provided.
JOURNAL OF PROCESS CONTROL
(2023)
Article
Engineering, Environmental
Abdul Samad, Iftikhar Ahmad, Manabu Kano, Hakan Caliskan
Summary: The use of an artificial intelligence model as a surrogate in the online optimization of process conditions of reactive units of a petroleum refinery under uncertainty improves the exergy efficiency of the process. An artificial neural network model, combined with genetic algorithm and particle swarm optimization, achieved efficient process optimization. Sensitivity analysis revealed that the inlet temperatures of reactors were the most influential variables affecting the process exergy efficiency.
PROCESS SAFETY AND ENVIRONMENTAL PROTECTION
(2023)
Article
Computer Science, Artificial Intelligence
Sayaka Ogawa, Koichi Fujiwara, Manabu Kano
Summary: False heart rate feedback can improve player experience. The most effective heart rate feedback pattern is to accelerate by 5bpm per minute.
IEEE TRANSACTIONS ON AFFECTIVE COMPUTING
(2023)
Article
Engineering, Chemical
Ryosuke Masuda, Yoshinari Hashimoto, Max Mulder, Marinus M. (Rene) van Paassen, Manabu Kano
Summary: In this study, an automatic control system for hot metal temperature (HMT) was developed to achieve accurate process control in a blast furnace. The control algorithm, based on a two-dimensional transient model and non-linear model predictive control (NMPC), successfully reduced the control deviation of HMT compared to conventional manual operation.
DIGITAL CHEMICAL ENGINEERING
(2023)
Article
Engineering, Biomedical
Koichi Fujiwara, Koshi Ota, Shota Saeda, Toshitaka Yamakawa, Takatomi Kubo, Aozora Yamamoto, Yuki Maruno, Manabu Kano
Summary: This study proposes a method for detecting symptoms of heat illness based on heart rate variability analysis. By monitoring abnormal changes in heart rate variability caused by heat stress, the method aims to prevent exacerbation of heat illness. The results of the experiment on 103 volunteers at risk of heat illness development showed a sensitivity of 75% and a false-positive rate of 1.02 times per hour. The proposed method will contribute to receiving appropriate treatment for heat illness before exacerbation, thereby protecting people's health.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Thermodynamics
Jihad Salah Khan, Iftikhar Ahmad, Usman Khan Jadoon, Abdul Samad, Husnain Saghir, Manabu Kano, Hakan Caliskan
Summary: This study proposes a gray-box modeling approach for predicting the optimum mass flow rates of inlet streams in a Plate and Fin Heat Exchanger under uncertainty. By integrating genetic algorithm with a first principle model and replacing it with an artificial neural networks model, a novel gray-box framework is developed, which exhibits better effectiveness and higher outlet temperature than the direct application of the first principle model. The performance of this gray-box model is comparable to the integrated framework of genetic algorithm and first principle model, but with significantly reduced computation time, enhancing the heat exchanger's energy recovery and robustness to cope with uncertainty.
INTERNATIONAL JOURNAL OF HEAT AND MASS TRANSFER
(2023)
Proceedings Paper
Automation & Control Systems
Shota Kato, Kazuki Kanegami, Manabu Kano
Summary: Digital twins are crucial for digital transformation, and physical models are essential for their realization. This study proposes an automated AI, named AutoPMoB, to facilitate the building of physical models. The study focuses on judging the equivalence of variable definitions, and introduces a method based on ProcessBERT that outperforms the original BERT and SciBERT methods in terms of accuracy.
Article
Chemistry, Medicinal
Keita Yaginuma, Shuichi Tanabe, Manabu Kano
Summary: This study evaluated three types of gray-box models in fluidized bed granulation and proposed an assessment method based on Hotelling's T-2 and Q residual, which contributes to decision support in selecting gray-box or white-box models.
CHEMICAL & PHARMACEUTICAL BULLETIN
(2022)
